Machine learning approaches to constrain and understand the role of clouds in climate change (ML4CLOUDS)

作为我们这个时代的一个决定性挑战，气候变化促成了2015年《巴黎协定》，其核心政策目标是将全球变暖控制在2摄氏度以下。然而，在实际气候变化预测方面仍然存在很大的不确定性，这意味着有一个很宽的日期窗口，在此期间可能会超过这一阈值。假设温室气体持续排放，这可能在未来十年内实现，也可能要到世纪后半叶。为了给决策者提供信息，决策者迫切需要减少这种不确定性。我们的研究项目ML 4CLOUDS旨在解决云在这种不确定性中的主导作用以及对气候变化的耦合影响。云是无处不在的现象，随时覆盖地球表面的三分之二，因此在我们的气候系统中发挥着关键作用。至关重要的是，在大气二氧化碳（CO2）浓度增加的情况下，云是全球变暖预测中最重要的不确定因素。云也是厄尔尼诺南方涛动（ENSO）等主要气候变化模式的关键调节器，而ENSO又会导致区域气候和天气极端事件。因此，更好地了解云的响应及其与大气环流和全球变暖的相互作用已被世界气候研究计划列为七大挑战之一。因此，限制与云相关的不确定性，并了解潜在的物理驱动因素，将对社会产生宝贵的价值。云的基本作用主要来自于它们与地球能量收支的相互作用。低空云层对阳光具有高度反射性（对气候具有冷却作用），而对流层上层的云层则捕获来自地球的辐射（具有变暖作用）。云的形成本身就向大气层释放潜热。正是这些过程对大气温度和水文循环的总体影响使云对气候系统的行为和演变如此重要。ML 4CLOUDS旨在更好地了解驱动云形成的复杂物理控制机制。这将提高我们预测地球云量在人类影响下如何变化的能力，例如增加大气二氧化碳和气溶胶污染，从而减少全球变暖的不确定性。这种与云相关的不确定性的减少也将反馈给我们建模和理解当今气候变化的能力，以及我们对ENSO等主要气候模式未来变化的预期。我们将通过一种结合人工智能的新方法实现这些目标这些方法结合有针对性的气候反馈分析和在超级计算机上运行的最先进的气候模型模拟。具体来说，我们的项目将：1。使用机器学习从大型气候模型数据集和天基观测中推导出云控制关系。这些关系将提供更好的云响应估计，并大大减少物理气候变化预测的不确定性。它们将进一步提供新的见解，以了解云响应背后不同物理机制的相对重要性。从观测中了解到的云控制关系也将有助于为未来的气候模型开发提供信息，例如新的英国地球系统模型（UK-ESM）。提高我们对云在调节气候变率主要模式中的作用的理解。除了对极端天气的重要性之外，气候变异性还叠加在人为气候变化造成的长期趋势上。因此，更好地了解云在气候变率中的作用将提高我们检测和归因历史气候变化的能力，并预测未来气候变化及其极端情况。

As a defining challenge of our time, climate change has led to the 2015 Paris Agreement whose central policy goal is to keep global warming well below 2 degrees Celsius. The substantial remaining uncertainty in physical climate change projections, however, means that there is a very wide window of the dates within which this threshold might be passed. Assuming continuous greenhouse gas emissions, it could be within the next decade, or it might not be until well into the second half of this century. To inform their decision-making, policymakers urgently need this uncertainty reduced. Our research proposal, ML4CLOUDS, addresses the leading role of clouds in this uncertainty, and the coupled implications for climate variability.Clouds are ubiquitous phenomena covering around two thirds of Earth's surface at any time and, as such, play key roles in our climate system. Crucially, clouds are the single most important uncertainty factor in global warming projections under increasing atmospheric carbon dioxide (CO2) concentrations. Clouds are also key modulators of the main modes of climate variability, such as the El Niño Southern Oscillation (ENSO), which in turn drive regional climate and weather extremes. A better understanding of the response of clouds and their interactions with the atmospheric circulation and global warming has therefore been highlighted as one of the 7 Grand Challenges by the World Climate Research Programme. Constraining cloud-related uncertainties, and understanding the underlying physical drivers, would consequently be invaluable to society.The fundamental role of clouds primarily arises from their interaction with Earth's energy budget. Low-altitude clouds are highly reflective for sunlight (having a cooling effect on climate), while upper tropospheric clouds trap radiation emitted from the Earth (having a warming effect). Cloud formation itself releases latent heat to the atmosphere. It is the overall impacts of these processes on atmospheric temperature and the hydrological cycle that make clouds so important for the behaviour and evolution of the climate system.ML4CLOUDS aims to provide a better understanding of the complex physical control mechanisms driving cloud formation. This will improve our ability to predict how Earth's cloud cover will change under human influences such as increasing atmospheric CO2 and aerosol pollution, and thus reduce uncertainty in global warming. This reduction in cloud-related uncertainty will also feed back on our ability to model and comprehend present-day climate variability, and on how we expect the main climate modes, such as ENSO, to change in the future.We will achieve these goals through a novel approach incorporating artificial intelligence (or machine learning) methods, paired with targeted climate feedback analyses and state-of-the-art climate model simulations run on supercomputers. Specifically, our project will:1. Use machine learning to derive cloud-controlling relationships from large climate model datasets and from space-based observations. These relationships will provide improved estimates of the cloud response and significantly reduced uncertainty in physical climate change projections. They will further provide new insights into the relative importance of distinct physical mechanisms behind the cloud response. Cloud-controlling relationships learned from observations will also be helpful to inform future climate model development, e.g. of the new UK Earth System Model (UK-ESM).2. Improve our understanding of the role of clouds in modulating the main modes of climate variability. Next to its importance for extreme weather, climate variability is superimposed on long-term trends due to man-made climate change. A better understanding of the role of clouds in climate variability will therefore enhance our ability to detect and attribute historical climate change, and to predict future changes in climate and its extremes.